Building

• Nomi — local-first AI coding agent. Plan review before execution, capability-gated tools, BYO LLM. Tauri desktop + Go daemon. Your repo never leaves the laptop.
• Roady — plan-of-record for AI coding agents. v0.12 shipped: nested sub-projects, live Kanban, cross-project board, drag-and-drop, SSE updates, auth gate.
• TokenOps — local daemon measuring LLM spend, rate-limit headroom across Claude Max / ChatGPT / Copilot / Cursor, and output-side prompt patterns. v0.17.
• Nox — open-source security scanner with first-class AI app security (prompt injection, embedding leakage, agent over-privilege, MCP hardening, AIBOM). Apache 2.0, Go. v0.10.
• Armada — campaign orchestration for Jira. Atlassian Marketplace, v4.2. Canonical site at armada.run.
• Fortify — resilience patterns for Go services calling LLMs (circuit breaker, retry, rate limit, hedge, cost budget, stream timeout). v1.5.
• Bolt — zero-allocation slog.Handler for Go with first-class OpenTelemetry. genai add-on for OTel GenAI semconv.
• Statekit — typed statechart engine for Go. Backend workflows + AI agent runtimes on the same primitives.
• Cognitive stack — mnemos (evidence + memory), chronos (time/pattern perception), nous (commitment extraction), praxis (typed action layer), olymp (runtime / control plane). Domain-agnostic Go pieces an agent runtime composes.
• mcp-go — Go framework for building MCP servers. Used by most of the projects above.

Writing on

• The plan-vs-evidence split in agent memory — why most "memory" features fix half the bug.
• The cost of skipping primitives in an agent runtime — what looks like a model problem usually isn't.
• Deploy / release / launch decoupling for teams shipping AI features.
• Field Notes essays land at /writing; subscribe in the footer.

Reading

• Team Topologies (re-read).
• The Goal (Goldratt) — flow-not-utilisation reframe.
• Click.
• Vibe Coding.
• Rethinking operating models.

Inspired by nownownow.com.